How to deploy neural network model



I have already trained and tested a neural network model. The results on the test data were good, so I am planning to deploy this model for future predictions.

My question is for how many periods can I generate predictions. For example, if I used 4 years of data for training and 1 year for testing, then for how many years/periods can I generate predictions?.




Hi @Nafeeza86 -

I suppose the answer is “it depends”. What kind of prediction are you doing? Is this a classification model, or forecasting with time series, or…?

If you are using 4 years of data for training and 1 for testing, regardless of the amount of time encompassed, they key question is this: how similar is your new, unknown data to the data you built your model with? You might have an idea about this up front, but you probably won’t be able to know for sure. So the modeling process will probably be iterative, and involve looking at how well predictions on your new dataset go, and how future predictions accuracy might (or might not ) decay over time.

Sorry that my answer lacks specifics. At any rate, start small with your predictions - maybe a few months or a year - and then extend your time horizon if the model performs well.




I am trying to build a prediction model for time series forecasting. My model predicts the next day hourly energy consumption. The inputs for the model are: month, hour of the day, day of the week, previous day consumption, etc.

My question is once I deploy the model to predict the next day’s energy consumption, do I have to include these predictions into my dataset and retrain the model to get the forecast for the other day and so forth on?

Also, what do you mean by unknown data? I am assuming the test set.

Thanks for your support!